[FFmpeg-cvslog] avutil/pixelutils: faster pixelutils_sad_[au]_16x16

[Clément Bœsch]

ffmpeg | branch: master | Clément Bœsch <u at pkh.me> | Thu Aug 14 22:30:55 2014 +0200| [45c7f3997ea11c3d1007b2126b1c0049a8c27105] | committer: Clément Bœsch

avutil/pixelutils: faster pixelutils_sad_[au]_16x16

~560 → ~500 decicycles

This is following the comments from Michael in
https://ffmpeg.org/pipermail/ffmpeg-devel/2014-August/160599.html

Using 2 registers for accumulator didn't help. On the other hand,
some re-ordering between the movs and psadbw allowed going ~538 to ~500.

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=45c7f3997ea11c3d1007b2126b1c0049a8c27105
---

 libavutil/x86/pixelutils.asm |   14 +++++++++-----
 1 file changed, 9 insertions(+), 5 deletions(-)

diff --git a/libavutil/x86/pixelutils.asm b/libavutil/x86/pixelutils.asm
index 8ab0a18..15213d9 100644
--- a/libavutil/x86/pixelutils.asm
+++ b/libavutil/x86/pixelutils.asm
@@ -134,16 +134,20 @@ cglobal pixelutils_sad_16x16, 4,4,5, src1, stride1, src2, stride2
 %macro SAD_XMM_16x16 1
 INIT_XMM sse2
 cglobal pixelutils_sad_%1_16x16, 4,4,3, src1, stride1, src2, stride2
-    pxor        m2, m2
-%rep 8
-    mov%1       m0, [src2q]
+    mov%1       m2, [src2q]
+    psadbw      m2, [src1q]
     mov%1       m1, [src2q + stride2q]
-    psadbw      m0, [src1q]
     psadbw      m1, [src1q + stride1q]
-    paddw       m2, m0
     paddw       m2, m1
+%rep 7
     lea         src1q, [src1q + 2*stride1q]
     lea         src2q, [src2q + 2*stride2q]
+    mov%1       m0, [src2q]
+    psadbw      m0, [src1q]
+    mov%1       m1, [src2q + stride2q]
+    psadbw      m1, [src1q + stride1q]
+    paddw       m2, m0
+    paddw       m2, m1
 %endrep
     movhlps     m0, m2
     paddw       m2, m0